Statistics 351 (Fall 2015)

Prof. Michael Kozdron

September 14, 2015

Lecture #3: Multivariate Random Variables

Let (X, Y )0 be a random vector. (With only two components it is traditional to use X and
Y instead of X1 and X2 .)
Sometimes we are interested in the distribution of just one component.
In the continuous case, we have
fX (x) =
and
FX (x) =

fX,Y (x, y) dy

fX (u) du =
1

x
1

fX,Y (x, y) dy.

We call fX (x) the marginal density (function) for X (or just marginal or just density) and
FX (x) the (marginal) distribution function for X. Similar formul hold for fY (y) and FY (y).
Definition. The random variables X and Y are independent if and only if
FX,Y (x, y) = FX (x) FY (y).
i.e., i fX,Y (x, y) = fX (x) fY (y) in the continuous case.
Exercise. Write down the definition of the marginal mass function and the marginal distribution function in the discrete case.
Example (Chapter 1, Exercise 1.2). Suppose that

X
0
(X, Y ) =
Y
denotes the coordinates of a dart thrown uniformly at random at a circular dart board. To
be specific, suppose that the circle is centred at the origin and has radius 1. We can describe
the random vector (X, Y )0 by its density function
(
1/, if x2 + y 2 1,
fX,Y (x, y) =
0,
otherwise.
Suppose that (X, Y )0 is a point uniformly distributed in the unit disk so that
(
1/, if x2 + y 2 1,
fX,Y (x, y) =
0,
otherwise.
Determine the distribution of X.
31

x2

Solution. By definition,

x2

Figure 1: The region {x2 + y 2 1}.

fX (x) =
=

That is, if x is fixed between

Therefore,

fX,Y (x, y) dy
1
p
1 x2
p

1 x2

easy step!

1
dy watch limits!

1 and 1, then y ranges between

fX (x) =

2p
1

x2 ,

1 x 1.

fY (y) =

2p
1

y2,

1 y 1.

Similarly,

x2 and

x2 .

We can now ask the following two questions.

Are X and Y independent?
Are X and Y uncorrelated?
Clearly, X and Y are NOT independent since fX (x) fY (y) does NOT equal fX,Y (x, y). It
turns out, however, that X and Y are uncorrelated.
Recall. If X, Y are random variables, then
Cov(X, Y ) = E(X

EX)(Y

Corr(X, Y ) = X,Y = p

EY ) = E(XY )

Cov(X, Y )
Var(X) Var(Y )

(EX)(EY ), and

Note that = X,Y is a scale-invariant real number with 1 1. Also note that in the
continuous case,
ZZ
E(XY ) =
xyfX,Y (x, y) dx dy.
R2

32

In order to show that X and Y are uncorrelated, we need to show that Cov(X, Y ) = E(XY )
(EX)(EY ) = 0. Since fX,Y (x, y) = 1/ for x2 + y 2 1, we have
ZZ
1
E(XY ) =
xy dx dy.

{x2 +y 2 1}

To compute this double integral, use polar coordinates: x = r cos , y = r sin , 0 r 1,

0 < 2, dx dy = r dr d. That is,
ZZ
Z Z
Z 2
1
1 1 2 3
1
xy dx dy =
r cos sin dr d =
cos sin dr d = 0.

0 0
4 0
{x2 +y 2 1}

Furthermore, we find
Z 1
2p
E(X) =
x
1

x2

dx = 0 and E(Y ) =

1
1

2p
1

y 2 dy = 0

recognizing that the integral of an odd function over a symmetric interval is 0. (Or, one can
compute the integrals via first-year calculus substitutions.)

33